PCM vs. DSD: Encoding Method Accuracy, SINAD, and What Actually Limits Modern DACs

[Chrispy]

Because Paul McGowan says it's the best.

It MUST be true.

LOL must have overlooked that nonsense!

 

[DWPress]

I have honestly never considered DSD, high resolution recordings or upsampling anything beyond 48khz to be anything to fuss about or waste money on. If you want to obsess about minuscule differences in digital format (that may be audible or not to a trained ear) that's time and money far better spent on better speakers or learning a bit about DSP.

Digital packaging makes no difference at all on 8" 2-way speakers but a bit of EQ to suite your listening space and maybe some judicious room treatment (how about a rug?) will make it sound much more like what the engineer and artist intended than how the 1's and 0's are arranged.

edit: all you headphone people may have strong opinions otherwise.

 

[Redacted]

Anecdotally some listeners claim to hear a difference between PCM or DSD playback of the same recording and anecdotally some claim no difference.

I see dead people. They are not usually in a decent band.

 

[Scytales]

The obvious problem with DSD is that you can't edit it.

Actually, you can.

The first commercially available DSD editing process is decipherable from the paper and drawings in its presentation made by A. Nishio (Sony) in 1997 at the 102nd AES Convention :

Obviously, this device let one DSD stream (designated 'A') or the other ('designated 'B') pass at the output, with only a delay (which is only a matter of filling or emptying a memory). This delays has the same length than the time required by the digital processing when a cross-fade is performed between DSD stream A and B. This is necessary in order for all streams to remains synchronized respective to each others.

When a cross-fade is performed, both strings of 1 bit data at the inputs are scaled in values by digital faders (the blocks that apply the scaling factor Ka and Kb).

At the faders, the data-streams becomes multibit. Thanks to the numerical space opened at this stage, the underlying information conveyed by the DSD 1 bit streams (i.e. a "1" means a pulse in the positive direction and a "0" means a pulse in the opposite direction), no longer merely is a directional command (up/down), but becomes a true integer capable of precisely expressing fractions of the maximum magnitude of the pulses.

One string of data is progressively scaled down by a continuously decreasing scaling factor (for instance x1, x0.95, x0.9, x0.85, etc...) whereas the other is progressively scaled up from a muted state by a continuously increasing scaling factor (for example x0.05, x0.1, x0.15, x0.2 etc). The two multi-bit streams are mixed together and the sum is re-modulated back to a new 1 bit stream that conveys the information contained in the summation signal.

One of the key feature of the process is the block called "Pattern Match Detector" (PMD). Switching between two 1 bit streams having two widely different patterns of successive 1s and 0s risks producing a transient noise, if only because the DC components between each stream would be very different. To avoid that, the PMD continuously compares the patterns of 0s and 1s in the respective processed and unprocessed 1 bit streams and orders to switch only at the moment when the patterns are sufficiently alike to avoid transient noise. This is somewhat akin a 0 cross detection in a stepped analogue volume control integrated circuit, where the switching between steps occurs only when the analogue signal crosses 0 to avoid transient noise.

The process as a whole is not very complex to understand; it has been used (and likely still is) to produce thousands of SACDs using digital audio workstations (DAWs) equipped with Sony CXD2926 processors (the first Sony/Sonic Solutions prototypes, first-generation Sony Sonoma, SADiE DSD8, Sonic Solutions DSD.1). I have never heard audiophiles complain that edit points—which are inevitable in music production—are noticeably audible when listening to SACDs.

However, the process devised by Sony is highly sensitive to the design of the Delta-Sigma modulator used for SACD production. Since then, Philips has developed a different cross-fade processing system between two 1 bit streams—more flexible than Sony's—which has been integrated into the Merging Pyramix system from around the year 2000.

A lot of water has flowed under the bridge since those two implementations, and I don't know if anything new has been designed.

 

[krabapple]

That kind of simple editing is hardly the only production step that audio requires. And you can't do those in DSD.

 

[john61ct]

HiRes FOMO - Am I really missing anything?

I think the best argument for higher sampling rates is to move the output filter way beyond the 20Khz region so there's no chance of hearing any roll off. My middle aged ears don't care, but some young reviewers have demonstrated an ability to hear roll off with certain DACs. You may ask...

www.audiosciencereview.com